NIKI: Neural Inverse Kinematics with Invertible Neural Networks for 3D HPS

This repo contains the code of our paper:

NIKI: Neural Inverse Kinematics with Invertible Neural Networks for 3D Human Pose and Shape Estimation

Jiefeng Li*, Siyuan Bian*, Qi Liu, Jiasheng Tang, Fan Wang, Cewu Lu

In CVPR 2023

Different Inverse Kinematics Framework

Installation instructions

# 1. Create a conda virtual environment.
conda create -n niki python=3.8 -y
conda activate niki

# 2. Install PyTorch
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.6 -c pytorch -c nvidia


# 3. Install PyTorch3D (Optional, only for visualization)
conda install -c fvcore -c iopath -c conda-forge fvcore iopath
conda install -c bottler nvidiacub
pip install git+ssh://[email protected]/facebookresearch/pytorch3d.git

pip install -r requirements.txt

# 4. Pull our code
git clone [email protected]:Jeff-sjtu/NIKI.git
cd NIKI

Models

• Download our pretrained model from [Google Drive | Baidu (code: z4iv)].

Fetch data

• Download parsed annotations from [Google Drive].

Train

python scripts/train.py --cfg configs/NIKI-ts.yaml --exp-id niki-ts

Evaluation

python scripts/validate.py --cfg configs/NIKI-ts.yaml --ckpt niki-ts.pth

Demo

Download pretrained HybrIK and Single-stage NIKI models from onedrive link, and put them in exp/ folder.

python scripts/demo.py --video-name {VIDEO-PATH} -out-dir {OUTPUT-DIR}

Citing

If our code helps your research, please consider citing the following paper:

@inproceedings{li2023niki,
    title     = {{NIKI}: Neural Inverse Kinematics with Invertible Neural Networks for 3D Human Pose and Shape Estimation},
    author    = {Li, Jiefeng and Bian, Siyuan and Liu, Qi and Tang, Jiasheng and Wang, Fan and Lu, Cewu},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
}